(off topic - Penguin) What's wrong and right with Linux. Nov 1, 1999
By Steven J. Vaughan-Nichols, Sm@rt Reseller
Can You Trust This Penguin?
Cheaper Than You Think
Linux's Holes
Fear Of Heights
Loose Links
What Else Did We Forget?
Desktop Linux?
Learning To Samba
The Bottom Line
Let's get real. we all know there's a ton of Linux hype out there.
A day does not go by without another vendor declaring Linux support.
But there's also a lot of fear, uncertainty and doubt about Linux.
In this exclusive report, Sm@rt Reseller cuts through the hoopla and
reveals where Linux plays best, and where its use should be avoided.
Some of our findings--based on our internal tests and interviews with
key Linux experts--may surprise you.
Certainly, Linux has momentum. Shipments of the operating system are
expected to increase 25 percent annually through 2003, predicts
International Data Corp. But those numbers can be misleading.
Linux has yet to take root on the desktop and lacks many high-end
features that Unix proponents take for granted.
Still, Linux has some clear strengths. With Samba for file serving and
Apache for Web serving, it has proven to be a robust solution for file,
print and Web servers. It's also an important operating system for
network appliances, whether they are simple file servers or complete
Internet solutions like Rebel.com's Net Winder OfficeServer, where
stability and remote administration are top requirements.
Conventional wisdom says Linux is incredibly stable. Always skeptical,
we decided to put that claim to the test over a 10-month period. In our
test, we ran Caldera Systems OpenLinux, Red Hat Linux, and Windows NT
Server 4.0 with Service Pack 3 on duplicate 100MHz Pentium systems with
64MB of memory. Ever since we first booted up our test systems in
January, network requests have been sent to each server in parallel for
standard Internet, file and print services. The results were quite
revealing. Our NT server crashed an average of once every six weeks.
Each failure took roughly 30 minutes to fix. That's not so bad, until
you consider that neither Linux server ever went down. This test,
coupled with our technical staff's extensive Linux and NT experience,
leads us to believe that Linux truly is more stable than NT on
uniprocessor servers.
Cheaper Than You Think
Linux also is cheap--but you may not realize how cheap. Install Linux on
a server, and all of your customers can access that server for free.
By contrast, nearly all other server operating systems require client
access licenses. Those licenses can cost anywhere from $30 to $40 per
desktop. Multiply that "per seat" license by 50 users, and suddenly
the price tag for a non-Linux environment jumps from $1,500 to $2,000.
Your customers can't beat Linux when it comes to hardware costs, either.
With Linux, any low-end Pentium server (heck, even many 486s) still can
be a workhorse. Try any of the other major network operating systems on
an aging Pentium box and see how far you get. Not even Microsoft has the
guts to begin a Windows 2000 discussion here. Although Linux is
considered a potential Windows killer, it's important to remember that
it's also a threat to Unix variants. There's a reason why IBM, SCO and
Sun Microsystems are moving quickly to Linux compatibility. All of those
vendors have no choice, since many Unix resellers are taking the Linux
plunge--at least on the low end.
Part of the reason for that is Linux already has many of Unix's
client-server and host-based applications. With almost all major DBMSes
now shipping, at least their database engines on Linux, the future looks
especially bright for e-commerce, vertical-market resellers and even
application service providers.
Linux's Holes
Even as major software developers embrace Linux, however, some office
applications will lag. For example, Lotus is porting Domino Server
(Notes' heart) to Linux; the Notes client for Linux is still a rumor.
It's also unclear how many, and when, Notes developers will port their
APIs to Linux. You must make sure that your customers' enterprise
programs, especially those that are NT-based, will be available on Linux
before doing the penguin hop.
Even if Linux clears the app hurdle, it stumbles when it comes to
high-end networking. That's because the OS lacks numerous advanced
features, such as a journaling file system (JFS). It also offers only
basic support for symmetric multiprocessing (SMP), clustering and
storage-area networks (SANs). Those holes should leave "true" Unix
to dominate the high-end market for the foreseeable future.
Virtually all of the Linux experts we conferred with for this article
agree that Linux needs a JFS, which would vastly improve reliability.
With a JFS, a file server that crashes after someone accidentally cuts
off the power (OK, we admit it has happened here) can be up and running
with its files intact and restored in a few minutes. If your customer
considers any length of downtime intolerable after a "normal" crash,
a JFS is a must.
JFS is an architecture that logs all changes made to the file system.
Introduced by IBM in AIX, JFS provides both additional speed and added
security for high-end server operating systems by enabling them to check
and restore files based on the operating system's own "memory" of the
correct file ordering.
JFS could find its way into Linux, under the guise of the ext3 file
system. Ext3, as blessed by Linus Torvalds himself, will be
upward-compatible with the current ext2 file system, but is a few months
from deployment. Silicon Graphics' open-source XFS file system also
could make the grade as a JFS for Linux. Still, don't expect to see
either of those file systems in commercial deployment until mid-2000,
at the earliest.
Fear Of Heights
Linux also has scalability limits. Though commercial releases
include SMP support, they don't cut the mustard for some customers.
"Linux has a hard time scaling much beyond two or four CPUs,"
concedes Aaron McKee, strategic products manager for TurboLinux Inc.
"These configurations are far more common than some may give credit to."
SMP improvements are coming. Erik Troan, director of engineering at Red
Hat, says the forthcoming Linux 2.4 kernel handles SMP "much, much
better." The new kernel will be available in source code this month,
but it won't find its way into commercial Linux releases until February.
Still, scalability is relative. On low-powered, uniprocessor servers,
Linux wipes the floor with both Windows NT and NetWare. We proved that
in at least two recent head-to-head product reviews.
The story is dramatically different on high-end SMP servers.
Here, NT outpaces Linux at file serving, and Solaris outraces
them both in Web serving.
The moral of the story? Customers that need multiple low-end servers
for basic networking likely will be quite satisfied with Linux.
If you choose to recommend the SMP path, NT currently beats Linux
for file serving, and Solaris outpaces both of them at Web serving.
Loose Links
Linux has at least two other shortcomings on the high end.
Clustering technology for the operating system remains in its infancy,
and SAN capabilities are, for the most part, nonexistent.
Just ask Open Source Initiative leader Eric Raymond.
"[Linux] clustering is still bleeding-edge experimental," he concedes.
TurboLinux has a proprietary clustering product that works well in
informal tests, but most Linux developers would like to see an
open-source solution with better fail-over support.
As is, clustering fail-over support and process migration still needs
work. For today, you should recommend and sell clustered Linux solutions
only if you have a clustering guru on staff. It's not for the faint of
heart or those new to Linux. If your customer demands clustering,
IBM's AIX is likely the better option.
Linux also has weaknesses on the SAN front. The operating system does
not score well when it comes to support for shared-storage options,
whether done with Fibre Channel or SCSI. Work is being done toward both
approaches, but you won't see results from that effort anytime this year.
A related problem is the use and management of logical storage volumes.
As long as the file system is relatively fixed to the physical drive,
neither Linux, nor any other operating system, is going to achieve
six-figure (99.9999 percent) reliability, which is the magic number for
many demanding e-commerce customers.
SGI's Irix file system, XFS, has a logical volume manager that might be
able to divorce logical drives from actual hard drives. In theory, that
will allow resellers to add and subtract physical drives without downtime.
However, XFS's logical volume manager may end up being a proprietary
value-add, according to SGI lead engineer Jeremy Allison.
If that is so, Linux developers will need to hack a solution on their own.
What Else Did We Forget?
In the Windows world, you can never have enough memory. But Torvalds
insists that 2GB of memory is all that an x86-based Linux ever will
demand. If you need more, he quips, there's always Alpha hardware,
which Red Hat Linux supports. In theory, that combo scales to 8GB.
Other Linux supporters, like McKee from TurboLinux, claim that Linux
for x86 needs a higher memory limit. "Some large Unix-based database
servers, for example, are typically outfitted with 4GB, 8GB, or even
more memory," he notes.
Large memory support isn't expected to debut in the Linux 2.4 kernel.
However, you can expect to see it by mid-2000.
Linux also has shortcomings on the administration front. While Computer
Associates and other systems-management vendors race to bring Linux into
the managed operating-system fold, more work needs to be done.
"There is a robust set of SNMP tools and capabilities with Linux,"
asserts TurboLinux's McKee. "[But] most of these tools have not been
tightly integrated into the operating system, and configuration can
be a bit unintuitive for new users."
Indeed, without some manual configuration, McKee says a Linux box
can appear as a black hole within most corporate networks that use
network-management applications, such as Hewlett-Packard's OpenView
and IBM's Tivoli .
Desktop Linux?
Notice how we've been concentrating on Linux as a server operating
system? There's good reason. Desktop Linux--despite the K Desktop
Environment (KDE) and Gnome graphical user interfaces (GUI)--still falls
short of the Windows and MacOS competition. Themes, by which the desktop
is made to look and feel like other operating systems or your customers'
own unique GUIs, help, but the real problems lie in the lack of solid
underlying tools.
Erik Troan, director of engineering for Red Hat, puts it well: "You
start with a lack of applications. Corel is working [to correct this]
with its complete Corel Office Suite, as is Sun with Star Office."
But today, Troan concedes, he "still has to use second-tier checkbook
applications, and we still need strong presentation software."
Why is that so? Because Linux lacks a unified desktop infrastructure
API, a la Windows 9x. "With nothing similar to Microsoft's Common Object
Model," says Troan, "it's harder to write end-user GUI applications."
Until Linux has a common object layer, writing desktop applications
will continue to be slow.
Enter Inprise's Delphi and Corel's forthcoming Kylix rapid application
development (RAD) for C, C++ and Delphi. Kylix also will implement the
Borland Visual Component Library (VCL), giving Linux an object-style
layer. For Windows or Mac developers who want to do a little work with
Linux, those are exactly the kind of packages needed to start porting
applications to Linux.
Meanwhile, Corel is leading the way to an end-user-friendly GUI.
Based on what we've seen, it looks like Corel could improve Linux's
GUI by the end of this year.
And for a business deployment, Corel's Linux and WordPerfect Suite
for Linux may be all the applications an office customer needs.
Whether any businesses will be willing to swap out Windows
and Microsoft Office for Corel's Linux alternatives remains to be seen.
Learning To Samba
Strictly speaking, Samba, the open-source Common Internet File System
(CIFS) file and print server, isn't a Linux product at all. Samba also
can be found providing Windows NT file and print services for FreeBSD,
SCO UnixWare and even Solaris (thanks to Veritas Software Corp.).
With Samba, almost any Unix machine can be transformed into an NT
server, at least from an end user's viewpoint. Samba also is fast.
In our recent review of Samba, we found that on modest single-processor
systems, Samba's NT services ran rings around NT's own services.
Samba isn't perfect, but most horror stories about its use are
exaggerated. Reports of corporations being halted in their tracks by
someone placing a Samba system in an NT network as a primary domain
controller (PDC) are completely misleading. Any NT network would come
to a howling halt with NT systems dueling for PDC superiority. As an
ordinary server or a backup domain controller (BDC), a Samba-equipped
Linux box works as smooth as silk on NT networks. In one Sm@rt Reseller
branch office, we've been running a pair of Samba boxes with NT PDC and
BDC boxes since March. During that time, we've never seen a hint of a
conflict.
Fast, stable and secure--so what needs improvement? Lots of things,
according to Jeremy Allison, a leading Samba developer and SGI engineer.
Support for Samba as a PDC is a major, ongoing issue, which is "'bloody
hard." NT printer-driver support is almost there, but almost only counts
with horseshoes and hand grenades--not printers.
Another "exciting" project is WinBind. When successful, WinBind will
replace Linux user and group authentication with NT PDC access.
In brief, Samba users will have only a single log-in for access to all
CIFS, NT, and Samba file and print services. Currently, on hybrid
NT/Unix networks, a user must log in to the Linux box and then log
in to the NT side to access all available drives.
Samba also currently lacks the ability to map NT access control lists
(ACLs) to Portable Operating System Interface for Unix (Posix) ACLs.
But, it can map NT ACLs to standard Unix permissions. For many potential
customers--especially federal government offices--Posix compliance will
make Samba solutions much more compelling.
As it sits, Samba only works with the ASCII character set. SGI's Allison
is working to build Unicode support into Samba. Unicode is an ASCII
superset used to represent up to 65,356 characters instead of ASCII's
256. That is important because Unicode is the character representation
set of choice for most non-Roman alphabets. Practically speaking, Samba
will become more viable for international customers after that change is
implemented.
Finally, the Samba team isn't just working on making Samba a better NT
than NT within its range. The group already is working on Windows 2000
"down- level" compatibility--the same level of Windows 2000
interoperability that NT 4 will have under Windows 2000 networks.
Face it: Thanks to Samba, Microsoft-style file-and-print serving is now
as ubiquitously available on Unix servers as it is on NT. Even better,
it's faster than NT on modest servers--and more stable.
Advantage: Samba.
Can You Trust This Penguin?
The Bottom Line
All things considered, Linux makes a great server operating system
from the low end up to middle tier. Assuming Linux can run the
applications your customers need, it's very hard to beat in those markets.
On the desktop and in the enterprise arena, however, Linux continues
to have its problems. And, despite the endless reams of hype, we don't
expect that to change anytime soon.
Saint Or Sinner?
Halos:
Cheap Fast Stable
Horns:
Limited SMP support
Lacks JFS/Advanced storage options
Limited clustering support
Total Cost Of Ownership (TCO)
Data: Cost of a 50-user Windows 9x network that accesses file
and print services via Caldera Linux or Windows NT.
Also, minimal server configuration assumed...
The following is where the home page is. zdnet.com
My only contact is a request for permission to post a specific article.
Note: Do not copy or reference this SI Doug A K post, per see below.
Copyright (c) 1999 ZDNet.
All rights reserved.
Reproduction in whole or in part in any form or medium
without express written permission of ZDNet is prohibited.
ZDNet and the ZDNet logo are trademarks of Ziff-Davis Inc. |
